A Software Tool for Automatic Generation of Neural Hardware

Natural neural networks greatly benefit from their parallel structure that makes them fault tolerant and fast in processing the inputs. Their artificial counterpart, artificial neural networks, proved difficult to implement in hardware where they could have a similar structure. Although, many circuits have been developed, they usually present problems regarding accuracy, are application specific, difficult to produce and difficult to adapt to new applications. It is expected that developing a software tool that allows automatic generation of neural hardware while using high accuracy solves this problem and make artificial neural networks a step closer to the natural version. This paper presents a tool to respond to this need: A software tool for automatic generation of neural hardware. The software gives the user freedom to specify the number of bits used in each part of the neural network and programs the selected FPGA with the network. The paper also presents tests to evaluate the accuracy of the implementation of an automatically built neural network against Matlab.